No, I'm starting to bore myself. So I thought instead that today I'd write about postmodernism, data science, and how the two intersect. I really love the concepts that come out of postmodernism. They changed how I view everything from how I read film to how I practice my religion.
But what does postmodernism mean for my work as a data scientist? Let's step back a moment.
Many decades ago postmodern theory changed the way that we understood what it meant to create or engage with art, literature, and culture. As a reaction to modernism, postmodernism rejected the notion that there was one pure, perfect way to create or understand something. People had grown tired of the dictatorial pursuit of soulless perfection characterized by many modern thinkers. The great Jacques Tati relentlessly mocked such ideas in his films like Mon Oncle:
Michael Lewis recently saw folks going into finance after reading his book on the financial crisis. His was a cautionary tale, but others saw it as a glimpse into the easy money that Wall Street offered the incompetent. Even though Lewis thought the book accurately portrayed Wall Street's awfulness, others saw the easy money and wanted in on the opportunity. Lewis said, "You never know what book you wrote until you know what book people read."
Postmodernism presents us with the idea of the writable text -- a text doesn't have one meaning, but rather, the reader is inherently an interpreter and can find meaning in the text other than what the author intended. This heightened view of context and individuality has long been ignored in business when dealing with customers in favor of monolithic solutions. It was ignored out of necessity, because a business could never understand individual context. But today, even a small business like MailChimp (where I work) can through data science.
Let me give an example from my own life.
In the past, Bayesian models presented a modern definition of spam -- there is some platonic notion of spam and through analysis of words and phrases in an email, we can make a determination once and for all whether something is spam.
But that's not true these days. A lot of spam (as the law defines it) is not about Viagra or Nigerian Princes -- it's just regular ol' marketing material for small businesses that lacks permission. In other words, the text can be interpreted many ways from a postmodern perspective, but what makes spam super-spammy is the interpretation of that mail by the recipients. Not the content itself -- the recipients' reading of that content.
So our models must now predict and interpret those readings. At MailChimp we do this via data -- we track all kinds of things about email addresses and then predict whether a list of emails wants content from a user or not. If the list were different, even for the same content, the prediction would change. That's the postmodern, reader-centric perspective of data science in action.
Postmodern literary theory introduced the concept of intertextuality -- that the meaning of a text is shaped by other texts through allusions, quotations, parody, homage, etc. This is to say that no one text stands in isolation with meaning unto itself, but rather everything lends meaning to everything else it touches in an intertextual web.
This is the bread and butter of data science. For businesses, their users, readers, customers, etc. on the internet are all texts to be read and understood. All of my interests, likes, posts, clicks are references that lend context and meaning to me. And these interactions connect me to millions of other humans on the internet. We then all become texts, each lending meaning to each other directly and indirectly as we create and interact with content online.
If I respond to your tweet, then naturally I'm referencing you, in which case you are in a way an intertextual reference when understanding the body of online content that makes up my digital presence.
And if we both buy the same juicer on Amazon, then by both referencing that juicer, we are lending context to each other. If you then buy an electric chain saw, your next purchase says something vague about me given that at one time our juicing interests aligned. Perhaps I too would like to buy an electric chain saw.
This type of intertextuality then gives rise to the data science practice of collaborative filtering. Making suggestions to a user based on the behaviors of those whom the user references.
Collaborative filtering moves us away from monolithic approaches to marketing where we believe all users want the same thing (perhaps causing us to design to the middle) and closer to a fluid, neotribalistic marketing approach where individuals are targeted with specific content based on how we read their intertextual presence.
Reactions -- Playfulness and Paranoia
Folks are going to handle this new way of viewing the world differently, and just like in postmodern literature, those responses seem to vacillate between playfulness and paranoia. In postmodern art, intertextual references are used for fun. A great example of this is Gilmore Girls, which exemplified the use of referential humor. Similarly in the world of data science, examining the interconnectedness of people and making decisions based on model outputs, is a blast. And you see individuals and companies toying with these ideas. Friend suggestion, social navigation, and sentiment analysis are all examples of play. People are toying with this intertextual data and are creating new and fun products out of it. Like Cleverbot, where the chat bot recycles past conversations with humans in its repertoire and throws them back at you as if the words were its own.
On the flip side, postmodern thought gives rise to the concept of paranoia. We cannot understand things holistically. This world is complex, chaotic, and any situation, person, text, etc. can be read in multiple ways. So then how do we respond when we discover secret ordering principles behind the world we once thought was un-understandable? The NSA is reading our mail and doing the dreaded 3 hop query. LinkedIn knows I just had a bowel movement. Amazon knows I want that blender. We respond (and rightfully so) just like Heller did in Catch-22 -- with paranoia.
I'd argue that a mixture of these emotions, excitement on one hand and fear on the other, is healthy. Those who end up too paranoid may fail to reap the benefits of the world these technologies are ushering us into. I for one love the collaborative filtering products offered by Netflix, Amazon, and Twitter. Their recommendations save me time and brain power that'd I'd rather spend blazing through episodes of Parks and Recreation.
On the other hand, those who get too excited ended up hurting people. They violate privacy and propriety in unexpected ways. Excitement can blind your critical eye regarding how using these new tools can marginalize people.
Shilling my book in the conclusion
The reason why I wrote my book was because I wanted to communicate this perspective on data to more people than a small class of data science practitioners. The explosion in transactional data storage and intellectual capital around acting on that data has shifted how businesses can engage with their customer base. It's possible for even a small company to now move from a modern way of thinking ("Everyone is the same. There is one strategy for dealing with customers") to a postmodern way of thinking ("Everyone is different. Context matters. I can use data to better communicate individually with folks"). And I wanted to help enable that shift. That's why I wrote chapters on using supervised AI to target certain customers and another on using unsupervised AI to detect communities of people in social graphs.
So buy my book. I packed it with intertextual humor -- hell, I've even got a 90s dance music playlist in there. And a chapter on Anduril, Flame of the West. It doesn't get more postmodern than that, does it?